Skip to main content
Human-in-the-Loop (HITL) in Agno enables you to implement patterns where human oversight and input are required during agent execution. This is crucial for:
  • Validating sensitive operations
  • Reviewing tool calls before execution
  • Gathering user input for decision-making
  • Managing external tool execution

​
Types of Human-in-the-Loop Flows

Agno supports four main types of human-in-the-loop flows:
  1. User Confirmation: Require explicit user approval before executing a tool
  2. User Input: Gather specific information from users during execution
  3. Dynamic User Input: Have the agent collect user input as it needs it
  4. External Tool Execution: Execute tools outside of the agent’s control
Currently Agno only supports Human-in flows for Agent. Team and Workflow will be supported in the near future!

​
Resolving Human-in-the-Loop Requirements

During Human-in-the-Loop flows, the agent run will pause to wait for the user to resolve whatever requirement is needed. This is how you can interact with the requirements in your code: 
# We run the Agent and get the run response
run_response = agent.run("Perform sensitive operation")

# In our run_response, we will find a list of active requirements:
for requirement in run_response.active_requirements:

    # We can now iterate over the requirements and resolve them:

    # For example, if the requirement needs user confirmation:
    if requirement.needs_confirmation:
        # Ask the user for confirmation
        confirmation = input(f"Do you approve the tool call to {requirement.tool.tool_name} with args {requirement.tool.tool_args}? (y/n): ")

        # Resolve the requirement by confirming or rejecting it, based on the user's input
        if confirmation.lower() == "y":
            requirement.confirm()
        else:
            requirement.reject()

​
Types of Human-in-the-Loop Requirements

There are three types of Human-in-the-Loop requirements you can handle in your code: User confirmation, user input and external tool execution. This is how you can check the type of a requirement in your code: 
for requirement in run_response.active_requirements:

    # If the requirement is about user confirmation:
    if requirement.needs_confirmation:
        ...

    # If the requirement is about obtaining user input:
    if requirement.needs_user_input:
        ...

    # If the requirement is about executing an external tool:
    if requirement.is_external_tool_execution:
        ...

​
Resuming Agent Execution

After all active requirements have been resolved, you will want to continue the run. You do this by calling the continue_run method: This is how the full flow would look like in your code: 
run_response = agent.run("Perform sensitive operation")

for requirement in run_response.active_requirements:
    # You handle any active requirements here

# After resolving all requirements, you can continue the run:
response = agent.continue_run(run_id=run_response.run_id, requirements=run_response.requirements)
The continue_run method continues with the state of the agent at the time of the pause. You can also call the continue_run method passing the RunOutput of the specific run to continue: 
response = agent.continue_run(run_response=run_response)

​
Streaming Human-in-the-Loop Flows

You can also stream the responses you get during a Human-in-the-Loop flow. This is useful when you want to process or show the response in real-time. It works similarly to non-streaming. You will just need to handle the events you get from streaming the Agent run. If any event is paused, then you will need to handle the active requirements: 
for run_event in agent.run("Perform sensitive operation", stream=True):
    if run_event.is_paused:
        for requirement in run_event.active_requirements:
            # You handle any active requirements here

    response = agent.continue_run(run_id=run_event.run_id, requirements=run_event.requirements, stream=True)
You can also stream the events resulting from calling the continue_run or acontinue_run methods.

​
Learn More

User Confirmation

Require explicit user approval before executing tool calls

User Input

Gather specific information from users during execution

Dynamic User Input

Let agents request user input dynamically when needed

External Tool Execution

Execute tools outside of the agent’s control

​
Developer Resources